cursor-agent

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Cursor Agent

Delegate a task to Cursor's headless CLI agent (

cursor-agent

). Use this to get a second opinion from a non-Claude model on code reviews, plan critiques, text/copy review, or general Q&A.

将任务委托给Cursor的无头CLI Agent（

cursor-agent

）。可用于在代码审查、规划评估、文本/文案审核或通用问答场景中，获取来自非Claude模型的第二意见。

Inputs

输入参数

```
$task
```
: The task to hand to cursor-agent. Usually a review or question prompt. If the user adds a model hint (e.g. "with Opus", "using GPT", "pomocí Sonnetu"), extract it and pass via
```
--model
```
.

```
$task
```
: 要交给cursor-agent的任务。通常是审查或问题提示。如果用户添加了模型提示（例如“with Opus”、“using GPT”、“pomocí Sonnetu”），请提取该提示并通过
```
--model
```
参数传递。

Goal

目标

Run cursor-agent headless, surface its output to the user, and add a brief Claude TL;DR so the user doesn't have to re-read a wall of text.

以无头模式运行cursor-agent，向用户展示其输出，并添加一段简短的Claude TL;DR（摘要），避免用户需要重新阅读大段文本。

Steps

步骤

1. Parse the task

1. 解析任务

Extract from

$task

and the surrounding conversation:

Model hint — look for phrasing like:
- "Opus" / "claude" →
```
--model claude-opus-4-7-medium
```
- "Sonnet" →
```
--model claude-4.6-sonnet-medium
```
- "GPT" / "GPT-5" / "Codex" →
```
--model gpt-5.3-codex
```
- "Gemini" →
```
--model gemini-3.1-pro
```
- "Auto" →
```
--model auto
```
- Nothing mentioned → omit
```
--model
```
  (cursor uses its default:
```
composer-2-fast
```
  )
Mode hint:
- "plan" / "plan mode" / "navrhni" →
```
--mode plan
```
- "write" / "edit" / "implement" / "headless" → omit
```
--mode
```
  (cursor can write)
- Otherwise →
```
--mode ask
```
  (read-only, default)
Is this a review? If the task mentions "review", "check", "verify", "second opinion", or references a branch/PR/diff, enrich the prompt (step 2).

Success criteria: Flags decided, prompt ready.

从

$task

及上下文对话中提取：

模型提示 — 查找以下表述：
- "Opus" / "claude" →
```
--model claude-opus-4-7-medium
```
- "Sonnet" →
```
--model claude-4.6-sonnet-medium
```
- "GPT" / "GPT-5" / "Codex" →
```
--model gpt-5.3-codex
```
- "Gemini" →
```
--model gemini-3.1-pro
```
- "Auto" →
```
--model auto
```
- 未提及任何模型 → 省略
```
--model
```
  参数（Cursor将使用默认模型：
```
composer-2-fast
```
  ）
模式提示：
- "plan" / "plan mode" / "navrhni" →
```
--mode plan
```
- "write" / "edit" / "implement" / "headless" → 省略
```
--mode
```
  参数（Cursor支持写入操作）
- 其他情况 →
```
--mode ask
```
  （只读模式，默认值）
是否为审查任务？ 如果任务中提到“review”、“check”、“verify”、“second opinion”，或引用了分支/PR/差异，请丰富提示内容（步骤2）。

成功标准：确定好参数，准备好提示语。

2. Enrich the prompt (reviews only)

2. 丰富提示内容（仅审查任务）

If the task is a code review:

Identify the diff scope. Try in order: explicit reference in the task (PR number, branch name), current branch vs
```
main
```
(
```
git diff origin/main...HEAD
```
), or uncommitted changes (
```
git diff HEAD
```
).
Prepend a short instruction to the prompt so cursor-agent knows what to look at:
"Review the branch
```
<name>
```
against main. Use
```
git diff origin/main...HEAD
```
to see the actual changes. Focus on real issues (correctness, edge cases, security); skip nitpicks. Report under ~400 words with
```
file:line
```
references."

For non-review tasks (ask/plan/text/UI), pass the task as-is.

Success criteria: Final prompt string ready for the CLI.

如果是代码审查任务：

确定差异范围。按以下顺序尝试：任务中明确引用的内容（PR编号、分支名称）、当前分支与
```
main
```
分支的差异（
```
git diff origin/main...HEAD
```
），或未提交的更改（
```
git diff HEAD
```
）。
在提示语前添加简短说明，让cursor-agent知道需要关注的内容：
"审查分支
```
<name>
```
与main分支的差异。使用
```
git diff origin/main...HEAD
```
查看实际更改。重点关注真实问题（正确性、边缘情况、安全性）；忽略琐碎细节。以不超过400字的篇幅报告，并附上
```
file:line
```
引用。"

对于非审查任务（问答/规划/文本/UI），直接传递原始任务即可。

成功标准：准备好用于CLI的最终提示字符串。

3. Run cursor-agent

3. 运行cursor-agent

Invoke headless via Bash:

bash

cursor-agent -p --output-format text [--mode <mode>] [--model <model>] "<prompt>"

Always use
```
-p
```
(print/headless).
Always use
```
--output-format text
```
unless the task needs structured output.
Workspace defaults to pwd — don't pass
```
--workspace
```
unless the user explicitly asked for a different dir.
For runs with only non-write modes (
```
--mode ask
```
or
```
--mode plan
```
), no extra permission flags are needed. For headless write mode, do not silently add
```
--force
```
or
```
--yolo
```
; ask the user first.
Pipe to
```
tail -n 200
```
if you suspect a very long response and want to cap context use. By default, take the full output.

Success criteria: cursor-agent returns (exit 0) with a response. If it errors (auth, model not available), surface the error to the user instead of hiding it.

通过Bash调用无头模式：

bash

cursor-agent -p --output-format text [--mode <mode>] [--model <model>] "<prompt>"

始终使用
```
-p
```
参数（打印/无头模式）。
始终使用
```
--output-format text
```
，除非任务需要结构化输出。
工作区默认为当前目录（pwd）——除非用户明确指定其他目录，否则不要传递
```
--workspace
```
参数。
对于仅使用非写入模式的运行（
```
--mode ask
```
或
```
--mode plan
```
），无需额外权限参数。对于无头写入模式，不要静默添加
```
--force
```
或
```
--yolo
```
参数；需先询问用户。
如果预计响应会很长，可通过管道传递给
```
tail -n 200
```
以限制上下文使用。默认情况下，保留完整输出。

成功标准：cursor-agent返回（退出码0）并给出响应。如果出现错误（授权问题、模型不可用），请向用户显示错误信息，不要隐藏。

4. Return the output

4. 返回输出

Present to the user in this order:

Cursor's verbatim response — quote it directly so the user hears the other model's voice, including file:line references.
Your short TL;DR — 3–6 bullets at most, separated by a horizontal rule. Highlight what's real vs. noise, and suggest concrete next actions. Do not silently filter cursor's findings; if you disagree with one, call it out explicitly.

Keep the summary tight — the goal is to save the user from re-reading, not to replace cursor's output.

Success criteria: User sees both cursor's raw output and a Claude-curated action list.

按以下顺序向用户展示内容：

Cursor的原始响应 — 直接引用，让用户听到其他模型的输出，包括
```
file:line
```
引用。
简短的TL;DR摘要 — 最多3-6个项目符号，用水平线分隔。突出真实问题与无关内容，并建议具体的后续操作。不要静默过滤Cursor的发现；如果您不同意其中某点，请明确指出。

保持摘要简洁——目的是节省用户重新阅读的时间，而非替代Cursor的输出。

成功标准：用户同时看到Cursor的原始输出和Claude整理的操作列表。

Notes

注意事项

```
cursor-agent --list-models
```
shows every available model ID. If the user asks for something exotic (e.g. "thinking tier", "max tier"), check the list first.
```
cursor-agent
```
reads/writes inside the current workspace like any CLI tool — it is not sandboxed by default. Treat its output the same way you'd treat a teammate's PR comment: trust but verify.
This skill is for one-shot delegation. For a longer back-and-forth, the user can run
```
cursor-agent
```
interactively themselves (no
```
-p
```
).

```
cursor-agent --list-models
```
命令会显示所有可用的模型ID。如果用户要求使用特殊模型（例如“thinking tier”、“max tier”），请先查看该列表。
```
cursor-agent
```
像其他CLI工具一样在当前工作区内读写——默认情况下未沙箱化。对待其输出应如同对待队友的PR评论：信任但需验证。
此技能适用于一次性任务委托。如需更长时间的交互，用户可自行以交互式方式运行
```
cursor-agent
```
（不使用
```
-p
```
参数）。